Probing the Puzzle of Paraguayan Temperatures

In the recent “Horrors of Homogenization” post I panned the warming adjustments applied by NCDC to the Paraguayan raw surface air temperature records, which are similar to those applied by GISS and BEST. However, further study suggests that these adjustments may – repeat may – not be quite so outrageous as they look, and in the interests of balance and objectivity this post presents a review of the Paraguayan records that addresses the question of whether they really do contain a cooling bias that would justify warming corrections. And the fact that the review fails to reach a conclusion serves as an example of how difficult it can sometimes be to make sense of conflicting temperature records in the same area.

We begin with a continental perspective. Figure 1 plots temperature trends since about 1950 (record length varies) at 90 selected South American stations. All those blue dots in Paraguay; are they caused by genuine cooling, or are they caused by Paraguay?

Figure 1:  Distribution of records showing warming and cooling, South America

A strong statistical case can be made that they are caused by Paraguay. Only 13 of the 80 records outside Paraguay (16.25%) show cooling, and if we apply this probability to Paraguay we find that the chances that almost all of the records there will show cooling are effectively zero. And because Paraguay’s boundaries do not follow any physiographic features that might be expected to impact climate there is no obvious reason why this probability should not be applied. (The only other South American country that shows net cooling since 1950 is Chile, but Chile is separated from the rest of the continent by the Andes.)

But when we zoom in on Paraguay and the surrounding area (Figure 2) we find that the correlation between cooling and Paraguay’s borders is not exact. Of the ten stations in Paraguay that have reasonably continuous records going back at least fifty years two (Puerto Stroessner and Pedro Juan Caballero) show warming, and Camiri and Yacuiba in Bolivia also show net cooling rather than warming:

Figure 2:  Distribution of records showing warming and cooling, Paraguay and surrounding area

What does the Paraguayan cooling look like? It’s defined by the eight congruous raw records shown in Figure 3, which include Puerto Stroessner even though it shows overall warming but exclude Encarnacion, which is an erratic record that matches none of the other records in the area (all the raw temperature data used in this post are downloaded from the BEST station list). A trend line drawn through the mean of all the records shows about a degree C of net cooling since 1940, with effectively all of it occurring between the early 1960s and the mid-1970s:

Figure 3:  Monthly means of Paraguayan records showing cooling (Asuncion, Puerto Casado, Pilar, San Juan Bautista, Puerto Stroessner, Concepcion, Mariscal, Bahia Negra. Mean of all records shown in black)

And what do the records from Argentina and Brazil that surround Paraguay look like? As shown in Figure 4 they are also reasonably congruous but collectively they show about six-tenths of a degree of gradual warming since 1940 (the records include the Paraguayan city of Pedro Juan Caballero, which connects seamlessly with the Brazilian city of Ponta Pora and is part of the same metro area):

Figure 4:  Monthly means of Argentinian & Brazilian records showing warming (Corrientes, Resistencia, Formosa, Posadas, Las Lomitas, Ponta Pora, Pedro Juan Caballero, Corumba. Mean of all records shown in black)

Figure 5 plots the difference between the means of the two sets of records shown in Figures 3 and 4. The peaks and troughs line up but the Paraguay records show a degree C of cooling relative to the Argentina-Brazil records between 1965 and ~1976.

Figure 5:  Comparison of means of Paraguay and Argentine/Brazil records shown in Figures 3 and 4

And now we come to the $64,000 question. Is this 10-year cooling period in (most of) Paraguay real or artificial? If it’s artificial it will have one of the following physical causations:

Station moves: BEST, which records these events in its data set, identifies no station moves in Paraguay between 1965 and 1976.

Time-of-observation changes: BEST identifies none of these either.

It is of course possible that there were unrecorded station moves or t_obs changes between 1965 and 1976, but if there were we should see accompanying stair-step shifts in the Paraguayan records that can be identified by subtracting them from the mean of the Argentina/Brazil records. When I applied this approach to the eight records shown in Figure 3, however, I identified only one such shift – in the Asuncion record at the end of 1966 – and it explains only about half of the total ~1.5C cooling differential between Asuncion and the Argentina/Brazil records since 1953:

Figure 6:  Asuncion record minus Argentina/Brazil mean (black line is running annual average)

Land use changes: These can cause gradual temperature shifts, but there is no evidence for any land-use changes in Paraguay over the 1965-76 period that were not replicated to at least some extent next door in Argentina and Brazil.

The only event I found that coincides in time with the 1965-1976 period is the tenure of Colonel Gumersindo Adolfo Da Silva as head of Paraguay’s Dirección de Meteorología e Hidrología between 1965 and 1977. Could the colonel have implemented changes over this period that led to nationwide cooling? No way of knowing.

And although the distribution of red dots in Figure 1 makes it look unlikely we also have to consider the possibility that the Paraguayan records are correct and that the Argentinian and Brazilian records are warming-biased– did I hear someone say “urban warming”? I looked into this question and here are the results. First a tabulation of station populations and trend-line warming/cooling gradients since ~1950 in degrees C/century:

And second an XY plot of the Table data:

Figure 7:  Warming/cooling trend versus population, all records

The data show no trend until we delete Asuncion, whereupon we see some evidence that urban warming could be having an effect, although the results are not exactly compelling. (One problem is that to evaluate urban warming impacts we need to know where the station was at all times in its history, but in many cases we don’t even know where the station is now. The incredibly precise lat-longs BEST gives for the Asuncion station – 25.34792 degrees south, 57.67917 degrees west – place it in a swamp over the border in Argentina.)

The only thing left to consider is the evidence that suggests that the Paraguayan cooling may be a real climatic effect. For what it’s worth, here it is:

1. One record in Paraguay – Pedro Juan Caballero – shows a warming gradient that matches the warming gradients in the adjacent Argentinian and Brazilian records. Another (Puerto Stroessner) shows overall warming. Two records in Argentina – Corrientes and Resistancia – show little or no warming.

2. Two adjacent records in Bolivia, Camiri and Yacuiba, also show cooling. The trend, however, is different to the cooling trend in Paraguay:

Figure 8:  Camiri (blue) and Yacuiba (red) monthly records versus Paraguay mean (black)

3. The large “hump” in the Paraguayan record between 2001 and 2004 is not seen in records outside Paraguay. If this feature is real – and it’s hard to see how it could be artificial – it indicates that things happen to the climate in Paraguay that don’t happen elsewhere. (Similar humps occur in some Amazon Basin records between 1960 and 1970.)

So where does all this leave us? It leaves us unable to say whether the Paraguayan records are cooling biased or not, and therefore unable to prove that the warming adjustments applied to them by BEST, GISS and NCDC are invalid (although if they are valid it’s for the wrong reasons; but that would have to be the subject of a separate post). How do we resolve this uncertainty, if in fact it can be resolved? And how do we treat the Paraguay records if it can’t?

Addendum:  Euan has suggested that it might be instructive to post a graph of temperature data from Chile, the only other country in South America which shows net cooling since 1950. The graph below plots the records for (from south to north) Punta Arenas, Isla Huafo, Puerto Montt, Valdivia, Isla Juan Fernandez, La Serena and Copiapo, with the mean again shown in black. Note that Punta Arenas and Copiapo are almost 3,000km apart.

Figure 9: Seven Chilean surface air temperature records

This entry was posted in Climate change and tagged , , , , , . Bookmark the permalink.

41 Responses to Probing the Puzzle of Paraguayan Temperatures

  1. Stan says:

    Clearly, the conclusion must be “There is no demonstrable reason to adjust the raw temps, thus the raw temps must stand.” Any adjustments are completely without any basis. The temps currently in the official temp record are therefore wrong.

    • It depends on whether you think raw temperature records are innocent until proven guilty, as I do, or guilty until rehabilitated through adjustments, as GISS, NCDC, CRU, BEST and some of our other commenters do.

  2. Stan says:

    Not much of a puzzle then. “Scientists” adjusted the temps to show a warming when there was none, and there was no valid basis whatsoever for the adjustments. The question then becomes “why”?

  3. mbe11 says:

    You do a lot of searching and comparsion but refuse to see the forest for the trees. The cooling also occurred in Chile so it is possible the cooling actually existed, there is actually no way to know. When you have thousands of records and than arbitarily change them the result is not a correction in the record, you have simply intraduced a bias of your own. It would be far better to simply throw all the records in the pot and come up with an average which over time would be as accurate as the noise in the records. You and Giss and the rest are simply putting a bias in the records to justify your political ideas of what is correct. Reminds me of the soialist ideas to reform society used by the communists and others, it is definitly not science despite all your work or theirs.

  4. mbe11 says:

    Stan checks to the heart of the matter with the word “Why”. Nobody knows the motivation of the staffers who made the adjustments. The motivation of the IPCC is clear. In their own words they want to redress the wrongs of the industrial west and help the third world recover from those wrongs by giving them money with a cut for them.

  5. Euan Mearns says:

    Roger, interesting post and nice maps – I need some tutoring.

    I think the Paraguay records need to be real. In that whole area you have a lot of neutral records and cooling records over the border in Argentina and Bolivia. Its interesting that the cooling in Chile starts about 10 years earlier. We had some similar time offsets in S Africa as well around Durban, you may recall. So perhaps it is the warm records to the S in Argentina that are anomalous? Are these perhaps urban stations?

    This whole area is interesting. The main chunk of Antarctica has congruous flat records. But when you come into the Antarctic Peninsula we see distinct warming that extends to S Georgia and S Sandwich. Pretty sure the warming in the Peninsula has to do with ocean currents (or change in albedo). And so I’d speculate that cooling in Chile has to do with ocean currents and that for some reason it took time to spill over Andes. I think the climate in this area is also dominated by competition between Pacific and Atlantic . There is a lot of interesting stuff going on.

    • Roger Andrews says:

      I think the Paraguay records need to be real.

      Having found a few more records that extend the cooling farther into Bolivia and also into the north tip of Argentina I’m beginning to agree.

      I’d speculate that cooling in Chile has to do with ocean currents and that for some reason it took time to spill over Andes.

      This is more than speculation. The Andes largely isolate Argentina and the rest of S America from the effects of the Peru-Chile (Humboldt) current, which is why temperature trends on either side of the crest line are different.

      Abrupt temperature trend changes over short distances are seen in other areas too. Trends north and south of the Himalayas are quite different, as we might expect. But there’s no ready explanation for why they should be so different on either side of the Davis Strait (Greenland warmed by about 3C relative to the Hudson Bay area between 1920 and 1930). The differences between the east and west Mediterranean, Nova Scotia and the area around the St. Lawrence river, west Mexico/Central America and the Caribbean, and southeast Australia and the Nullarbor plain also have no obvious explanation. So I guess there’s no reason why we shouldn’t have another break around Paraguay.

  6. William says:

    Euan has suggested that it might be instructive to post a graph of temperature data from Chile, the only other country in South America which shows net cooling since 1950.

    Is there something special about 1950? If you go back another 10 years the mean returns to zero. Why not post Chile since 1940 or even 1910:

    I added Antofagasta and Puahuel (near Santiago), though they make little difference.

    • William: I used 1950 as the start date because some of the Paraguayan records don’t have data before then.

      I brought up your plot of Chile data. It shows effectively no warming over the last 100 years.

      • William says:

        Thanks for in-lining the plot. Can I insert an tag myself or is it an admin thing?

        As you say, it looks dead flat for 100 years – that sounds more neutral than “net cooling since 1950”, but maybe that is just me.

        • The easiest way would be for you to insert a tag on your original graphic and repost it. Then I can bring it up.

        • William says:

          “dead flat for 100 years – that sounds more neutral”

          Although rather obviously untrue, as there are two large lumps, one up, one down. No trend over 100 years is perhaps better.

    • Euan Mearns says:

      It’s a nice chart William. I have a post tomorrow on S S America and was going to add this in at the end if that’s OK? Well I’m going to do it anyway 😉 Its part of an interesting story that runs from Paraguay, through Patagonia to the Antarctic Peninsula and into the main Antarctic continent.

      Only Roger and I can post charts this way. Though I see a growing number of WP blogs display charts when you post a URL – need to get a plugin.

    • A C Osborn says:

      Interesting how the 1940 and 1950 peaks echo the Paraguay temps that Paul Homewood found in the Raw data.

  7. A C Osborn says:

    How can they “adjust” the data of a whole country based on the “regional” trends?
    There are around 17 Stations with no trends either way, in addition to the stations cooling.

    There is also another serious factor to be taken in to consideration when looking at UHI, it is no good looking at the current population to see the warming impact on trend, you have to know the “Increasing Trend in Population” as well.
    Faster Growth causes faster UHI, A city that has been over 50,000 for decades will have most of their UHI decades ago and the increase will have slowed down, unless the urban sprawl suddenly surrounds the Station.
    I have already pointed out the work of Roy Spencer that shows that UHI starts with the first building and piece of Concrete/Paving etc and builds from there.

  8. Rud Istvan says:

    Roger and Euan, I studied this for months, after ‘scooping up’ lots of skeptical blog examples. Wrote essay When Data Isn’t, with many valid examples and footnotes. Some conclusions:
    a. There are at least three separate sets of land record issues. (1) Station move, instrument change, TOBS,… In theory ( but not practice) correctable from metadata and calibration, as in Iceland. (2) UHI. But this not as simple as urban/rural, as (California was my example) it corrolates with population and land use, not just ‘cities’. (3) Microsite issues, which can be anywhere and crop up at any time. The US project is instructional.
    b. None of the ‘adjustment/homogenization’ algorithms can adequately fix what is fundamentally not fit for climate purpose. NCDC GHCN is not neutral; for all stations with records back to 1900 (n= 162 when deliberately undersampling US), 2/3 are warmed by cooling the past. BEST ‘regional expectation QC erroneously warmed 166900. Menne stitching introduces a past cooling bias into regular station maintenance.
    c. The error bars around land measurement are grossly understated.
    d. The error bars around land/sea must be even more uncertain, since trade route biased with major ocean surfaces under/not sampled.
    The GISS notion that the GMST uncertainty is +/- 0.1C is ludicrous on its face. As you show here.

    • Euan Mearns says:

      it corrolates with population and land use, not just ‘cities’.

      This is the tree I am barking up. Its a complex issue. But when I start scanning around on Google Earth I’m astounded at how Man has reworked the Earth’s surface so much in some areas. I will float another idea that irrigation may also be an important variable.

      • Roger Andrews says:

        But how do we evaluate how much impact irrigation and land use changes have had on global temperatures? To do this we need detailed historic records of irrigation and land use changes, and we don’t have them.

        The same goes for station moves, t_obs and equipment changes. As I mentioned in the post we don’t even know where many stations are now, leave alone where they used to be. There are no records of t_obs and equipment changes over most of the world either.

        As for microsite changes, such as the bushes at Low Head Lighthouse, we have nothing remotely resembling the data we would need to evaluate global impacts at this level of detail. Nor the time for that matter.

        I think we should quit fixating on all these possible sources of error, which we are never going to be able to quantify adequately, and verify surface air temperature records by simple congruity checks, which are generally an effective way of confirming (or not) that records are showing real climatic trends. And what about urban warming? Well, congruity checks usually do a good job of identifying that too.

  9. Nick Stokes says:

    It seems to me that there is a fairly obvious explanation right under your noses. At Paul Homewood’s blog, commenter Eliza said
    “My father set up/fixed to specified standrads all the stevensons boxes stations for the WMO in Paraguay during the 70’s (1964-1976). He was an ardent non-believer in AGW from day 1, but retired in 1977”

    She thinks there is a warmist conspiracy there. But it seems that the stations were, about that time, transferred from Navy control to a civilian authority, and there may indeed have been a major fixup/installation of Stevenson screens. Stevenson screens are intended to exclude stray radiation, of which there would be plenty in Paraguay. Indeed, there was a major drop in the graph above of Paraguay temperatures around then.

    It seems to me likely that, as intended, the homogenisation algorithm picked up these changes and corrected, by making earlier temperatures cooler. And indeed, the unadjusted cooling trend is a Paraguay artifact.

    • JerryC says:

      Pure speculation.

    • A C Osborn says:

      Yes and there was a major drop in Chile as well, but ofcourse he wasn’t working there.
      Talk about clutching at straws, you still haven’t satisfactorily explained Iceland adjustments yet, despite all your attempts.

      • William says:

        A major drop? You mean the one that results from omitting data before 1950 (instead of plotting from 1940 or 1910, as I did above)?

        • A C Osborn says:

          Interesting that you should jump in with both feet before reading what was being discussed and get completely the wrong impression.
          What was being discussed was the CHANGES to the data made by GISS who removed the peaks in the 1950s& 60s in Paraguay which was originally identified by Paul Homewood.
          Nick Stokes was suggesting that it was a Legitimate change due to the changeover to Stephenson Screens during the period 1964-1976 which might have caused the DROP of about 1.5 degrees in temperatures from the highs of the 1950s & 60s .

          It has absolutely nothing to do with the OVERALL TREND.
          I was only suggesting that YOUR Chile data showed similar rapid drops in Temperature around the same period.

    • BEST records four station moves at three Paraguayan stations between 1964 and 1976 – at Asuncion, Puerto Casado and San Juan Bautista in February 1971 and at Asuncion in June 1972. It’s reasonable to assume that if Stevenson screens had been installed it would have been during these station moves.

      The graph below superimposes these four moves on the temperature records for these three stations.

      All three records show the “Paraguayan cooling” beginning some years before the station moves. There’s also no evidence for stair-step downward shifts in the records at the time of the station moves, which is what we would expect to see if they had generated artificial cooling. These results indicate that station moves and/or the installation of Stevenson screens probably did not cause the Paraguayan cooling and that the “unadjusted cooling trend” is probably not a “Paraguay artifact”.

      It seems to me likely that, as intended, the homogenisation algorithm picked up these changes and corrected, by making earlier temperatures cooler.

      The “corrections” to the Paraguay records are actually generated by BEST’s warming-biased algorithm. I hope to have a post up on this shortly.

      • A C Osborn says:

        Except BEST got it wrong about one or more of those station moves, it was dealt with over on Paul Homewood’s Forum, something to do with the WMO if I remember correctly.

      • A C Osborn says:

        It was not originally BEST that Paul was looking at but GISS, who had no Station data other than the values.

      • Euan Mearns says:

        What are the dots you are plotting here Roger?

        Did you follow any of the discussion on Climate Etc? You may want to read some of Mosh’s comments like this one:

        The notion that a global product should be used for local issues OR evaluated by looking at local issues fundamentally misunderstands what the product is meant to do.

        • The dots are running annual averages of monthly means. You can’t pick anything out of the raw monthly data.

          Mosh’s comment isn’t relevant to your post, which dealt with a local issue (Australia), but I think the basic question is whether a “global product” that averages two degrees C of warming in central Asia with no warming in the Antarctic and large temperature spikes in some areas with no spikes in others and lots of warming in the NH with not very much in the SH means anything. The more I look at it the more convinced I become that it doesn’t.

      • Nick Stokes says:

        But did BEST get the timing of those moves from metadata or from their own inference from the data?

        Each plot does show a fairly radical drop in 1965-72. It may be a real climate drop, but could also be from the screen program. It seems that the GHCN algorithm interpreted at least some of them to be artificial. Hard to say decisively they are wrong.

        • But did BEST get the timing of those moves from metadata or from their own inference from the data?

          It looks like they got it from metadata. “Empirical breaks” are shown separately:

          Each plot does show a fairly radical drop in 1965-72. It may be a real climate drop, but could also be from the screen program. It seems that the GHCN algorithm interpreted at least some of them to be artificial. Hard to say decisively they are wrong.

          That’s what I said in the post. However, evidence is beginning to accumulate that the cooling probably is real. The montage you just posted over on the South America thread adds a little more:

      • Sam Taylor says:

        Best publish their source code. I assume that you’ll be going through it and highlighting the precise parts of their algorithms which would lead to a warning bias. Perhaps giving an example on a synthetic dataset.

        That’s all it would take.

  10. Pingback: The Hunt for Global Warming: South America | Energy Matters

  11. manicbeancounter says:

    How do 8/10 of the records in Paraguay contain cooling? Roger Andrews and Nick Stokes both have a similar explanation – it was human intervention that has corrupted the records.
    In my opinion the most possible explanation is something else. From Wikipedia’s article on Climatology is explanation of the nature of climate and the modelling of it.

    Climate research is made difficult by the large scale, long time periods, and complex processes which govern climate. Climate is governed by physical laws which can be expressed as differential equations. These equations are coupled and nonlinear, so that approximate solutions are obtained by using numerical methods to create global climate models. Climate is sometimes modeled as a stochastic process but this is generally accepted as an approximation to processes that are otherwise too complicated to analyze.

    Climate is partly chaotic. There could be genuine sub-regional trends that are thrown up all over the place that contradict the global picture. Similarly, in some regions global trends can be amplified (The Arctic in c.1910-1940 and c.1975-2005) or be none-existent (Central Australia). How you approach the problem depends on the objective. There are two main objectives.
    First is create a global land temperature trend from irregularly and placed temperature stations. Short-term local factors should be removed to obtain the best approximation possible.
    Second – and for me the more important – is to determine how much of recent warming is human caused. In this case removing the local factors will give a false view of the explanatory power of any global model. This might particularly true if regional amplifications are removed, or if evidence of an early twentieth century warming period is adjusted downwards.

    • Euan Mearns says:

      Kevin, I think you’ll find that Roger is doing a bit of Devil’s advocacy here and that you’ll find he does believe the Paraguay records are good, just that he can’t be certain. In his professional life Roger would have told his employer that he couldn’t be certain and recommend that more data needed to be acquired – which in mining means drilling more wells, cutting more core and doing more geochemistry. In Paraguay we can’t easily get more data.

      • Euan: It’s not devil’s advocacy, it’s objectivity, a commodity which is sorely lacking in the climate debate these days.

        I go where the data take me. If they don’t support the AGW position, which I find to be the case most of the time, I say so. But if they do I say that too. And when I find there are two sides to a story and that I’ve only presented one I hasten to present the other. That’s what I’ve done, or at least tried to do, in this post.

        • manicbeancounter says:

          Thanks for the comment Roger. I entirely agree with your attitude. I chose the name “manicbeancounter” for my blog because I have found that by looking at data from different perspectives you get different answers. By comparing and contrasting the different perspectives you get the best answer – or you end up not knowing which is perspective is better, requiring further evaluation.
          In the case of Paraguay I am now in two minds. If you have Excel 2010 try this experiment.
          1. Copy Figures 3 and 4 onto a blank sheet.
          2. Stretch Fig. 4 so that the scales match up.
          3. Move Fig 4 over Fig 3, without releasing the image. You will see both images overlaid.

          – For the period 1940-1965 the averages are pretty much identical. The trend is pretty much the same.
          – For the period 1940-1965 the averages are pretty much identical. The trend is pretty much the same. The exception is that in Fig 3 there is a spike around 2003 of over twice the size of that in Fig 4.
          – It looks at first that about half the drop in the Paraguay data from 1965-1971 is in a single year. This is a conclusion I reached last month. However, in 1970-1971 there is a similar drop in the Argentina & Brazil data. But for 1965-1969 the Paraguayan series average falls, whilst Fig 4 rises.
          So, on the basis of the 2003 anomaly, the fall in the data might be a naturally-caused, local impact that cannot be currently explained. But on the basis of all the other data, and the near identical data with the Argentina & Brazil data for over 60 out of 70+ years, it seems some “adjustment” to the data has occurred.
          This is why I am now in two minds. However, there are a couple of audit checks that I hope to perform to verify either way.

          • Thanks manic:

            I think the experiment you propose is analogous to subtracting individual Paraguay records from the mean of the surrounding records in Argentina and Brazil, which I showed an example of in Figure 6. This approach should identify any large abrupt shifts in the Paraguay records relative to the Argentina/Brazil records if there are any, but all I found was the one shown in Figure 6, suggesting that the Paraguay cooling was gradual and therefore more likely to be natural than artificial. Using the same approach BEST identifies one other downshift in the Paraguayan records (at Mariscal) although it claims to have identified a lot more than that.

  12. Pingback: The Worst of BEST | Energy Matters

Comments are closed.